(is_non_overlapping_and_dense) gso to guard_or_false in when checking length 1 #158894
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
… length-1 & expected stride [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/158894
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (3 Unrelated Failures)As of commit 79fb24c with merge base 255a04b ( BROKEN TRUNK - The following jobs failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
UNSTABLE - The following job is marked as unstable, possibly due to flakiness on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
…en checking length-1 & expected stride" Switch from `guard_size_oblivious` to `guard_or_false` if you encounter a DDE, this would then fallback to computing elementwise strides. https://github.com/pytorch/pytorch/blob/2dccff7dcf56b0d168ebfd7ca08bdeca37273c56/torch/_prims/__init__.py#L1919-L1923 We think it's safe because Laith tested whether this fallback would fail any tests. It did not. #158157 ## Data-dependent exceptions (DDE) ``` File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 2139, in _to_copy x_tensor = torch._prims.convert_element_type(x_tensor, dtype) ... File "/data/users/colinpeppler/pytorch/torch/_prims/__init__.py", line 1920, in _convert_element_type_meta if torch._prims_common.is_non_overlapping_and_dense(a): File "/data/users/colinpeppler/pytorch/torch/_prims_common/__init__.py", line 494, in is_non_overlapping_and_dense if guard_size_oblivious(length == 1): GuardOnDataDependentSymNode: Could not guard on data-dependent expression Eq(u0 - 4, 1) (unhinted: Eq(u0 - 4, 1)). (Size-like symbols: u0) ``` ``` File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 2139, in _to_copy x_tensor = torch._prims.convert_element_type(x_tensor, dtype) ... File "/data/users/colinpeppler/pytorch/torch/_prims/__init__.py", line 1920, in _convert_element_type_meta if torch._prims_common.is_non_overlapping_and_dense(a): File "/data/users/colinpeppler/pytorch/torch/_prims_common/__init__.py", line 500, in is_non_overlapping_and_dense if guard_size_oblivious(stride != expected_stride): Caused by: unbacked_size.int(), # test/inductor/test_unbacked_symints.py:538 in fn (_prims_common/__init__.py:500 in is_non_overlapping_and_dense) ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov coconutruben [ghstack-poisoned]
…en checking length-1 & expected stride" Switch from `guard_size_oblivious` to `guard_or_false` if you encounter a DDE, this would then fallback to computing elementwise strides. https://github.com/pytorch/pytorch/blob/2dccff7dcf56b0d168ebfd7ca08bdeca37273c56/torch/_prims/__init__.py#L1919-L1923 We think it's safe because Laith tested whether this fallback would fail any tests. It did not. #158157 ## Data-dependent exceptions (DDE) ``` File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 2139, in _to_copy x_tensor = torch._prims.convert_element_type(x_tensor, dtype) ... File "/data/users/colinpeppler/pytorch/torch/_prims/__init__.py", line 1920, in _convert_element_type_meta if torch._prims_common.is_non_overlapping_and_dense(a): File "/data/users/colinpeppler/pytorch/torch/_prims_common/__init__.py", line 494, in is_non_overlapping_and_dense if guard_size_oblivious(length == 1): GuardOnDataDependentSymNode: Could not guard on data-dependent expression Eq(u0 - 4, 1) (unhinted: Eq(u0 - 4, 1)). (Size-like symbols: u0) ``` ``` File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 2139, in _to_copy x_tensor = torch._prims.convert_element_type(x_tensor, dtype) ... File "/data/users/colinpeppler/pytorch/torch/_prims/__init__.py", line 1920, in _convert_element_type_meta if torch._prims_common.is_non_overlapping_and_dense(a): File "/data/users/colinpeppler/pytorch/torch/_prims_common/__init__.py", line 500, in is_non_overlapping_and_dense if guard_size_oblivious(stride != expected_stride): Caused by: unbacked_size.int(), # test/inductor/test_unbacked_symints.py:538 in fn (_prims_common/__init__.py:500 in is_non_overlapping_and_dense) ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov coconutruben [ghstack-poisoned]
…en checking length-1 & expected stride" Switch from `guard_size_oblivious` to `guard_or_false` if you encounter a DDE, this would then fallback to computing elementwise strides. https://github.com/pytorch/pytorch/blob/2dccff7dcf56b0d168ebfd7ca08bdeca37273c56/torch/_prims/__init__.py#L1919-L1923 We think it's safe because Laith tested whether this fallback would fail any tests. It did not. #158157 ## Data-dependent exceptions (DDE) ``` File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 2139, in _to_copy x_tensor = torch._prims.convert_element_type(x_tensor, dtype) ... File "/data/users/colinpeppler/pytorch/torch/_prims/__init__.py", line 1920, in _convert_element_type_meta if torch._prims_common.is_non_overlapping_and_dense(a): File "/data/users/colinpeppler/pytorch/torch/_prims_common/__init__.py", line 494, in is_non_overlapping_and_dense if guard_size_oblivious(length == 1): GuardOnDataDependentSymNode: Could not guard on data-dependent expression Eq(u0 - 4, 1) (unhinted: Eq(u0 - 4, 1)). (Size-like symbols: u0) ``` ``` File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 2139, in _to_copy x_tensor = torch._prims.convert_element_type(x_tensor, dtype) ... File "/data/users/colinpeppler/pytorch/torch/_prims/__init__.py", line 1920, in _convert_element_type_meta if torch._prims_common.is_non_overlapping_and_dense(a): File "/data/users/colinpeppler/pytorch/torch/_prims_common/__init__.py", line 500, in is_non_overlapping_and_dense if guard_size_oblivious(stride != expected_stride): Caused by: unbacked_size.int(), # test/inductor/test_unbacked_symints.py:538 in fn (_prims_common/__init__.py:500 in is_non_overlapping_and_dense) ``` cc ezyang penguinwu bobrenjc93 voznesenskym EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov coconutruben [ghstack-poisoned]
ezyang
reviewed
Jul 23, 2025
torch/_prims_common/__init__.py
Outdated
| continue | ||
|
|
||
| if guard_size_oblivious(stride != expected_stride): | ||
| if guard_or_false(stride != expected_stride): |
There was a problem hiding this comment.
NGL, this is pretty suspicious. Let's suppose I have a tensor with size (u0, u1) and stride (u2, u3). If we somehow get to this loop, we will never trigger any of the conditions here, and then we will return True (that it is non overlapping and dense). But... it's clearly NOT??? Like I can trivially fill in values of u0..u3 that would make it overlapping. It is MUCH safer to guard_or_true here.
There was a problem hiding this comment.
That's a good point! I originally did not see it this way. This will be a better test case.
There was a problem hiding this comment.
Okay, so I decided not to replace gso here. Otherwise, we'd go into the compute_elementwise_output_strides path which means we'd need to address sorting unbacked strides.
…en checking length-1 & expected stride" Switch from `guard_size_oblivious` to `guard_or_false` if you encounter a DDE, this would then fallback to computing elementwise strides. https://github.com/pytorch/pytorch/blob/2dccff7dcf56b0d168ebfd7ca08bdeca37273c56/torch/_prims/__init__.py#L1919-L1923 We think it's safe because Laith tested whether this fallback would fail any tests. It did not. #158157 ## Data-dependent exceptions (DDE) ``` File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 2139, in _to_copy x_tensor = torch._prims.convert_element_type(x_tensor, dtype) ... File "/data/users/colinpeppler/pytorch/torch/_prims/__init__.py", line 1920, in _convert_element_type_meta if torch._prims_common.is_non_overlapping_and_dense(a): File "/data/users/colinpeppler/pytorch/torch/_prims_common/__init__.py", line 494, in is_non_overlapping_and_dense if guard_size_oblivious(length == 1): GuardOnDataDependentSymNode: Could not guard on data-dependent expression Eq(u0 - 4, 1) (unhinted: Eq(u0 - 4, 1)). (Size-like symbols: u0) ``` ``` File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 2139, in _to_copy x_tensor = torch._prims.convert_element_type(x_tensor, dtype) ... File "/data/users/colinpeppler/pytorch/torch/_prims/__init__.py", line 1920, in _convert_element_type_meta if torch._prims_common.is_non_overlapping_and_dense(a): File "/data/users/colinpeppler/pytorch/torch/_prims_common/__init__.py", line 500, in is_non_overlapping_and_dense if guard_size_oblivious(stride != expected_stride): Caused by: unbacked_size.int(), # test/inductor/test_unbacked_symints.py:538 in fn (_prims_common/__init__.py:500 in is_non_overlapping_and_dense) ``` cc ezyang penguinwu bobrenjc93 voznesenskym EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy chenyang78 kadeng muchulee8 amjames chauhang aakhundov coconutruben [ghstack-poisoned]
ColinPeppler
changed the title
pianpwk
approved these changes
Jul 25, 2025
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
|
I will approve to unblock current usecase, but I would probably rewrite this by introducing is_non_overlapping_and_dense_or_false and use it in the _convert_element_type_meta, that would be the unbacked semantics of that part. |
laithsakka
approved these changes
Jul 25, 2025
laithsakka
reviewed
Jul 25, 2025
|
I think i have an idea how to get rid of the size oblivious here, the main idea is the following: Answer : we would return False ! lol so not the end of the world if the function definitions is that said we should try to do correct sorting as much as we can using the Modulus idea. |
pytorchmergebot
pushed a commit
that referenced
this pull request
Jul 30, 2025
…3d bmm into 2d mm (#159184) Switch from guard_size_oblivious to guard_or_false if you encounter a DDE, this would then avoid folding this 3d bmm into a mm. https://github.com/pytorch/pytorch/blob/806d9e3fe70ec250a1fb3823841d16c61b7d1b02/torch/_decomp/decompositions.py#L4506-L4512 ## DDE ``` File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 4506, in matmul elif should_fold(tensor1, tensor2, is_out): File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 4472, in should_fold if guard_size_oblivious(t1.numel() == 0): torch.fx.experimental.symbolic_shapes.GuardOnDataDependentSymNode: Could not guard on data-dependent expression Eq(12*((u0//2)), 0) (unhinted: Eq(12*((u0//2)), 0)). (Size-like symbols: none) Caused by: (_decomp/decompositions.py:4472 in should_fold) ``` ``` File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 4506, in matmul elif should_fold(tensor1, tensor2, is_out): File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 4483, in should_fold return all( torch.fx.experimental.symbolic_shapes.GuardOnDataDependentSymNode: Could not guard on data-dependent expression Eq(3*((u0//2)), 3) (unhinted: Eq(3*((u0//2)), 3)). (Size-like symbols: none) Caused by: (_decomp/decompositions.py:4483 in should_fold) ``` Pull Request resolved: #159184 Approved by: https://github.com/ezyang ghstack dependencies: #158894
yangw-dev
pushed a commit
that referenced
this pull request
Aug 1, 2025
… length 1 (#158894) Switch from `guard_size_oblivious` to `guard_or_false` if you encounter a DDE, this would then fallback to computing elementwise strides. https://github.com/pytorch/pytorch/blob/2dccff7dcf56b0d168ebfd7ca08bdeca37273c56/torch/_prims/__init__.py#L1919-L1923 We think it's safe because Laith tested whether this fallback would fail any tests. It did not. #158157 ## Data-dependent exceptions (DDE) ``` File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 2139, in _to_copy x_tensor = torch._prims.convert_element_type(x_tensor, dtype) ... File "/data/users/colinpeppler/pytorch/torch/_prims/__init__.py", line 1920, in _convert_element_type_meta if torch._prims_common.is_non_overlapping_and_dense(a): File "/data/users/colinpeppler/pytorch/torch/_prims_common/__init__.py", line 494, in is_non_overlapping_and_dense if guard_size_oblivious(length == 1): GuardOnDataDependentSymNode: Could not guard on data-dependent expression Eq(u0 - 4, 1) (unhinted: Eq(u0 - 4, 1)). (Size-like symbols: u0) ``` Pull Request resolved: #158894 Approved by: https://github.com/pianpwk, https://github.com/laithsakka
yangw-dev
pushed a commit
that referenced
this pull request
Aug 1, 2025
…3d bmm into 2d mm (#159184) Switch from guard_size_oblivious to guard_or_false if you encounter a DDE, this would then avoid folding this 3d bmm into a mm. https://github.com/pytorch/pytorch/blob/806d9e3fe70ec250a1fb3823841d16c61b7d1b02/torch/_decomp/decompositions.py#L4506-L4512 ## DDE ``` File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 4506, in matmul elif should_fold(tensor1, tensor2, is_out): File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 4472, in should_fold if guard_size_oblivious(t1.numel() == 0): torch.fx.experimental.symbolic_shapes.GuardOnDataDependentSymNode: Could not guard on data-dependent expression Eq(12*((u0//2)), 0) (unhinted: Eq(12*((u0//2)), 0)). (Size-like symbols: none) Caused by: (_decomp/decompositions.py:4472 in should_fold) ``` ``` File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 4506, in matmul elif should_fold(tensor1, tensor2, is_out): File "/data/users/colinpeppler/pytorch/torch/_decomp/decompositions.py", line 4483, in should_fold return all( torch.fx.experimental.symbolic_shapes.GuardOnDataDependentSymNode: Could not guard on data-dependent expression Eq(3*((u0//2)), 3) (unhinted: Eq(3*((u0//2)), 3)). (Size-like symbols: none) Caused by: (_decomp/decompositions.py:4483 in should_fold) ``` Pull Request resolved: #159184 Approved by: https://github.com/ezyang ghstack dependencies: #158894
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Switch from
guard_size_oblivioustoguard_or_falseif you encounter a DDE, this would then fallback to computing elementwise strides.pytorch/torch/_prims/__init__.py
Lines 1919 to 1923 in 2dccff7
We think it's safe because Laith tested whether this fallback would fail any tests. It did not.
#158157
Data-dependent exceptions (DDE)
Stack from ghstack (oldest at bottom):
cc @ezyang @penguinwu @bobrenjc93 @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov @coconutruben